Information processing method, information processing device, production system, article manufacturing method, program, and recording medium

Abstract

To provide a method for allowing creation of a failure prediction model with high prediction accuracy even if only a small number of failure examples for a machine plant can be collected.SOLUTION: A control method includes steps of: acquiring a measured value related to the state of a machine plant from a normal state up to an abnormal state of the machine plant; extracting a plurality of feature quantities by using the measured value; acquiring, for each of the plurality of feature quantities, the maximum value of the degree of separation for separating the normal state and the abnormal state from each other and the timing at which the degree of separation becomes the maximum value; selecting a feature quantity used for machine learning based on the maximum value for each of the plurality of feature quantities; setting a period during which the machine plant has been in the normal state based on a timing in the selected feature quantity; extracting data of the selected feature quantity in the period as data for machine learning; creating a learned model through machine learning using the data for machine learning; and determining the state of the machine plant by using the learned model.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an information processing method, an information processing device, a production system, a program, and a computer-readable recording medium used to detect signs of mechanical equipment failure based on data from sensors, etc. In particular, the present invention relates to the creation of learning data used to generate a mechanical equipment failure prediction model by machine learning. [Background technology]

[0002] The operating state of machinery and equipment can change from moment to moment due to changes in the condition of its components, etc. If we say that an operating state within the allowable range in light of the intended use of the machinery and equipment is in a normal state, and an operating state outside the allowable range is in a faulty state, then for example, in the case of a production machine, a faulty state will result in the production of defective products or the halting of the production line, or other problems.

[0003] In production machinery, etc., it is common to perform maintenance work on a regular or irregular basis, even when the same work is performed repeatedly and continuously, in order to minimize the occurrence of breakdowns. Shortening the intervals between maintenance work is effective in increasing preventive safety, but because production machinery is stopped during maintenance work, excessively frequent maintenance work reduces the availability of production machinery. Therefore, it is desirable to be able to detect when a breakdown is imminent, even if the machinery is still operating normally. If it is possible to detect an imminent breakdown (predict a breakdown), maintenance work can be carried out at that point, preventing an unnecessary drop in availability.

[0004] A known method for predicting the occurrence of failures is to create a trained model in advance by machine learning the state of the machinery and equipment, and then use the trained model to evaluate the state of the machinery and equipment at the time of evaluation.To improve prediction accuracy, it is important to build a trained model that is suitable for predicting failures, but to do so, the suitability of the training data used in machine learning becomes an issue.

[0005] For example, Patent Document 1 describes extracting vectors based on sensor signals indicating the state of machinery and equipment, and selecting features to be used based on data checks of the feature vectors. It also describes creating a normal model of machinery and equipment based on the selected training data.

[0006] Furthermore, Patent Document 2 describes that, with regard to the selection of multiple learning data prepared for each season according to seasonal fluctuations, etc., the sensor signals to be focused on and selected according to the anomaly are selected based on the anomaly measure, which is the result of multivariate analysis, and the evaluation results of the influence of each sensor signal. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] Japanese Patent Application Laid-Open No. 2011-70635 [Patent Document 2] Japanese Patent Application Laid-Open No. 2011-59790 Summary of the Invention [Problem to be solved by the invention]

[0008] In mechanical equipment, measurement data is collected for various parameters to manage its operating status, but since the frequency of failures is generally low, it is not easy to collect data on many failure cases. On the other hand, the huge amount of measurement data collected for various parameters contains a mixture of data related to the occurrence of failures and data unrelated to them, making it redundant as learning data for machine learning.

[0009] Although Patent Documents 1 and 2 recognize the need to select learning data, they do not fully consider how to actually select and acquire learning data.

[0010] In particular, in the case of machinery and equipment where the frequency of failures is low and only a small number of failure cases can be collected, it has been difficult to create a trained model with high prediction accuracy using conventional methods. Therefore, there was a need for a method that could create a failure prediction model with high prediction accuracy even if only a small number of failure cases for mechanical equipment could be collected. [Means for solving the problem]

[0011] According to one aspect of the present invention, there is provided an information processing method for an information processing device having a processing unit, the processing unit acquiring at least two types of measurement values ​​relating to a state of the mechanical equipment as the mechanical equipment changes from a first state to a second state, acquiring an index for each of the at least two types of measurement values ​​when the mechanical equipment changes from the first state to the second state, and calculating an index from the at least two types of measurement values ​​based on the index. The specified measurement value choose In the predetermined measurement value, a period is set based on the timing when the maximum value of the index in the predetermined period is equal to or greater than a threshold value, and data for the period is This is an information processing method characterized by being used for machine learning.

[0012] According to another aspect of the present invention, there is provided an information processing device having a processing unit, the processing unit acquiring at least two types of measurement values ​​relating to a state of the mechanical equipment as the mechanical equipment changes from a first state to a second state, acquiring an index for each of the at least two types of measurement values ​​when the mechanical equipment changes from the first state to the second state, and calculating a value from the at least two types of measurement values ​​based on the index. The specified measurement value choose, A period is set based on the timing when the maximum value of the index in the predetermined period becomes equal to or greater than a threshold value, and data for the period is This is an information processing device characterized by using the above for machine learning. [Effects of the Invention]

[0013] According to the present invention, it is possible to select feature quantities that are highly correlated with the occurrence of a failure from among various feature quantities that indicate the state of mechanical equipment, and to create a failure prediction model with high prediction accuracy even if only a small number of failure cases are collected. [Brief explanation of the drawings]

[0014] [Figure 1]1 is a schematic functional block diagram for explaining functional blocks included in a failure prediction system according to an embodiment; [Figure 2] FIG. 2 is a schematic diagram illustrating a hardware configuration of the embodiment. [Figure 3] 5A and 5B are schematic diagrams for explaining a feature extraction method according to the embodiment. [Figure 4] 5A and 5B are schematic diagrams for explaining a method for calculating a degree of separation in the embodiment. [Figure 5] 1 is a schematic diagram illustrating a machine learning method according to an embodiment. [Figure 6] 1 is a schematic diagram for explaining a failure prediction method according to an embodiment; [Figure 7] 4 is a flowchart showing a processing procedure for generating a failure prediction model according to an embodiment. [Figure 8] 4 is a flowchart showing a procedure for failure prediction according to an embodiment. [Figure 9] FIG. 4 is a diagram showing an example of a work support screen displayed in the embodiment. [Figure 10] FIG. 10 is a diagram showing another example of the work support screen displayed in the embodiment. [Figure 11] 5A and 5B are schematic diagrams for explaining a method for determining a determination threshold value in the embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0015] With reference to the drawings, embodiments of the present invention will be described, including a failure prediction system, a control method, a control device, a machine equipped with a control device, a control program, a computer-readable recording medium, etc., used to predict failures in mechanical equipment. In the following description, an operating state within an allowable range in light of the intended use of the mechanical equipment may be referred to as a normal state, and an operating state outside the allowable range may be referred to as a failure state or an abnormal state.

[0016] [Function block configuration] FIG. 1 is a schematic functional block diagram for explaining the configuration of functional blocks included in a failure prediction system according to an embodiment. While FIG. 1 uses functional blocks to represent functional elements necessary for explaining the features of this embodiment, general functional elements that are not directly related to the problem-solving principles of the present invention are omitted. Furthermore, the functional elements shown in FIG. 1 are conceptual functional elements and do not necessarily have to be physically configured as shown. For example, the specific form of distribution and integration of the functional blocks is not limited to the illustrated example, and all or part of them can be functionally or physically distributed and integrated in any unit depending on the usage situation, etc.

[0017] As shown in FIG. 1, the failure prediction system of the embodiment includes a mechanical equipment 10 to be diagnosed and a failure prediction device 100. The machinery 10 is various types of industrial equipment, such as an articulated robot or production equipment installed on a production line. The machinery 10 is equipped with various sensors 11 for measuring the state of the machinery. For example, if the machinery 10 is an articulated robot, sensors for measuring the current value of the motors that drive the joints, joint angle sensors, and sensors for measuring speed, vibration, and sound may be installed. However, this is merely an example, and an appropriate number and type of sensors may be installed as the sensors 11 in appropriate positions depending on the type of machinery 10 and the purpose of the work. The sensors 11 may include various types of sensors, such as force sensors, torque sensors, vibration sensors, sound sensors, imaging sensors, distance sensors, temperature sensors, humidity sensors, flow sensors, pH sensors, pressure sensors, viscosity sensors, and gas sensors. While FIG. 1 illustrates the sensor 11 singularly for ease of illustration, multiple sensors are typically installed.

[0018] The mechanical equipment 10 is connected to the failure prediction device 100 via wire or wirelessly so as to be able to communicate with the failure prediction device 100, and the failure prediction device 100 can acquire data measured by the sensor 11 via communication. In the failure prediction model creation stage, the failure prediction device 100 uses data collected from the sensor 11 to select feature quantities highly correlated with the occurrence of a failure in mechanical equipment, and performs machine learning using the selected feature quantities to generate and store a trained model (failure prediction model). In addition, in the evaluation stage (failure prediction stage), the data collected from the sensor 11 at the time of evaluation is input into the trained model, and the degree of deviation is calculated using the input and output of the trained model to determine whether a failure is approaching. Below, the functional blocks of the failure prediction device 100 will be explained in order.

[0019] The failure prediction device 100 includes a control unit 110 , a storage unit 120 , a feature quantity selection unit 131 , a display unit 130 , and an input unit 140 . The control unit 110 and the feature selection unit 131 include multiple functional blocks, which are configured, for example, by the CPU of the failure prediction device 100 reading and executing a control program stored in a storage device. Alternatively, some or all of the functional blocks may be configured by hardware such as an ASIC provided in the failure prediction device 100.

[0020] The storage unit 120 includes a sensor data storage means 121, a feature storage means 122, a degree of separation storage means 123, a failure prediction model condition storage means 124, and a failure prediction model storage means 125. These means included in the storage unit 120 are configured by being appropriately allocated to storage areas of a storage device such as a hard disk drive, RAM, or ROM. The storage unit 120 is a data acquisition unit that acquires and stores various types of data required for processing to perform failure prediction.

[0021] The display unit 130 and the input unit 140 are user interfaces provided in the failure prediction device 100. The display unit 130 uses a display device such as a liquid crystal display or an organic EL display, and the input unit 140 uses an input device such as a keyboard, a jog dial, a mouse, a pointing device, or a voice input device.

[0022] The sensor data collection means 111 of the control unit 110 acquires measurement data from the sensors 11 of the mechanical equipment 10 and stores it in the sensor data storage means 121. That is, the sensor data collection means 111 collects and stores measurement data relating to the state of the mechanical equipment 10, such as the current, speed, pressure, vibration, sound, temperature of each part, etc., measured in the mechanical equipment 10.

[0023] The feature extraction means 112 extracts feature values ​​that indicate characteristics of the operating state of the mechanical equipment 10 based on the measurement data stored in the sensor data storage means 121 and stores the feature values ​​in the feature storage means 122. For example, the feature data may be the maximum and / or minimum values ​​of sensor measurement values ​​collected during one operating cycle of the mechanical equipment, or the average value. Alternatively, the feature data may be, for example, a value obtained by integrally transforming sensor values ​​over a predetermined period into a time-series frequency domain. The feature data may also be a differential value or a second-order differential value with respect to time of the sensor values ​​arranged in a time series. Furthermore, if the sensor measurement values ​​(raw data) themselves are useful as information for determining whether a failure is approaching, the measurement values ​​themselves may be treated as feature data. In this embodiment, the feature extraction means 112 extracts or calculates feature values ​​based on the sensor measurement data, creates time-series feature data, and stores the time-series feature data in the feature storage means 122.

[0024] The separation degree calculation means 113 acquires the feature quantities from the feature quantity storage means 122, calculates the maximum separation degree for each feature quantity, and stores it in the separation degree storage means 123. The separation degree, which will be described in detail later, is an index that indicates whether or not a feature quantity is sensitive to a change in state when the mechanical equipment changes from a normal state to a fault state.

[0025] The feature selection unit 131 acquires the maximum value of the degree of separation for each feature from the separation degree storage means 123, and selects the feature to be used for machine learning based on the acquired information and the threshold stored in the separation degree threshold setting means 132. Furthermore, the normal interval setting means 133 sets a normal interval of the mechanical equipment (a period of normal state in which signs of failure have not yet appeared in the feature) by referring to the timing at which the maximum value of the degree of separation for the selected feature appears. The selected feature and normal interval are stored in the failure prediction model condition storage means 124 as conditions for extracting learning data for machine learning.

[0026] When generating a trained model, the data extraction means 114 extracts machine learning data from the feature storage means 122 based on the conditions stored in the failure prediction model condition storage means 124, and outputs the machine learning data to the failure prediction model generation means 115. Furthermore, when evaluating, the data extraction means 114 extracts feature quantities of the same type as the feature quantities stored in the failure prediction model condition storage means 124 (features used for machine learning) from the feature quantity data at the time of evaluation stored in the feature storage means 122, and outputs the extracted feature quantities to the failure determination means 116.

[0027] The failure prediction model generation means 115 generates a trained model (failure prediction model) using the machine learning data input from the data extraction means 114 when generating the trained model, and stores the trained model in the failure prediction model storage means 125.

[0028] The fault determination means 116 inputs the feature data for evaluation input from the data extraction means 114 at the time of evaluation into a trained model (fault prediction model) stored in the fault prediction model storage means 125, calculates the degree of deviation between the input and the output, and then compares the degree of deviation with a determination threshold to determine whether or not there is a sign of a fault. The fault notification means 117 notifies an external device of the determination result of the fault determination means 116 and displays it on the display unit 130 .

[0029] [Hardware configuration] FIG. 2 shows a schematic diagram of an example of the hardware configuration of a failure prediction system according to an embodiment. As shown in FIG. 2, the failure prediction system may include PC hardware equipped with a CPU 1601 as a main control unit, a ROM 1602 as a storage device, and a RAM 1603. The ROM 1602 can store information such as processing programs and inference algorithms for implementing the failure prediction method described below. The RAM 1603 is used as a work area for the CPU 1601 when executing the control procedure. An external storage device 1606 is also connected to the control system. The external storage device 1606 may be an HDD, an SSD, or an external storage device of another system mounted via a network.

[0030] The processing program of the CPU 1601 for implementing the failure prediction method of this embodiment, which will be described later, can be stored in a storage unit such as an external storage device 1606 consisting of an HDD, SSD, etc., or in the ROM 1602 (e.g., an EEPROM area). In this case, the processing program of the CPU 1601 for implementing the failure prediction method can be supplied to each of the above storage units via a network interface (NIF) 1607 and updated with a new (different) program. Alternatively, the processing program of the CPU 1601 for implementing the failure prediction method can be supplied to each of the above storage units via various storage means such as magnetic disks, optical disks, and flash memories and their associated drive devices and updated. Various storage means, storage units, or storage devices storing a program capable of executing the processing of the CPU 1601 for implementing the failure prediction method constitute a computer-readable recording medium storing the failure prediction procedure of the present invention.

[0031] The sensor 11 shown in Fig. 1 is connected to the CPU 1601. In Fig. 2, for the sake of simplicity, the sensor 11 is shown as being directly connected to the CPU 1601, but it may be connected via, for example, IEEE488 (so-called GPIB) or the like. The sensor 11 may also be configured to be connected to the CPU 1601 via a network interface 1607 and a network 1608.

[0032] The network interface 1607 can be configured using a communication standard for wired communication such as IEEE 802.3 or wireless communication such as IEEE 802.11 or 802.15. The CPU 1601 can communicate with other devices 1104 and 1121 via the network interface 1607. For example, if the target of failure prediction is a robot, the devices 1104 and 1121 may be a supervisory control device such as a PLC or sequencer arranged to control and manage the robot, or a management server.

[0033] In the example shown in Fig. 2, an operation unit 1604 and a display device 1605 related to the input unit 140 and display unit 130 shown in Fig. 1 are connected as a UI device (user interface device). The operation unit 1604 can be configured by a terminal such as a handy terminal, or devices such as a keyboard, a jog dial, a mouse, a pointing device, and a voice input device (or a control terminal equipped with these). The display device 1605 may be any device that can display information related to the processing executed by the separation degree calculation means 113, the failure prediction model generation means 115, the failure determination means 116, etc. on a display screen, and may be, for example, a liquid crystal display device.

[0034] [About failure prediction methods] In this embodiment, the failure prediction model generation means 115 of the failure prediction device 100 constructs a trained model (failure prediction model) by so-called unsupervised learning. To learn the characteristics of mechanical equipment failures by so-called unsupervised learning, machine learning is performed using only operational data from a fault-free state, i.e., when the mechanical equipment is operating normally. Unsupervised learning is a method in which a learning device is provided with a large amount of input data alone, and the device learns the distribution of the input data. In other words, this is a method in which a device that performs processes such as compression, classification, and shaping on input data learns the process without providing it with training output data corresponding to the input data.

[0035] A failure prediction method using unsupervised learning techniques will now be described in detail. Machine learning is performed using only operational data of the machinery in its normal state, with the normal state being defined as a state in which the machinery is operating within an acceptable range in light of its intended use. In this embodiment, an autoencoder is used as the unsupervised learning model.

[0036] This embodiment is characterized by the method of extracting data used for machine learning, in which feature values ​​that are sensitive to state changes when mechanical equipment changes from a normal state to a faulty state are selected, and data for a specified period (normal section) of the selected feature values ​​is used as learning data.

[0037] First, feature extraction will be described with reference to FIG. 3. As shown in FIG. 3, mechanical equipment 10 to be subjected to failure prediction includes sensors 11, including sensor 1, a current sensor, sensor 2, a speed sensor, and sensor 3, a pressure sensor. In the failure prediction method according to this embodiment, feature values ​​indicative of the operating state of mechanical equipment 10 are extracted based on the measurement data of each sensor included in sensor 11. The feature values ​​are extracted by performing processes on the time-series measurement data of the sensors, such as integral transformation into the frequency domain, calculation of first and second derivatives with respect to time, filtering, and extraction of maximum and minimum values ​​of periodic behavior. The sensors and measurement data processing described above are merely examples, and any sensor or measurement data processing may be used as long as it can acquire data suitable for understanding the state of the mechanical equipment. Furthermore, if the state of the mechanical equipment can be easily analyzed using the sensor measurement data itself, the measurement data itself may be used as feature values ​​without any special processing. Figure 3 shows a schematic diagram of the state in which three types of processing are performed on the measurement data from Sensor 1, Sensor 2, and Sensor 3, and time-series data for a total of nine types of features are extracted, as shown on the right.

[0038] Next, a method for selecting, from the extracted feature quantities, feature quantities that change sensitively when the mechanical equipment changes from a normal state to a faulty state will be described. First, time-series data of the features of the mechanical equipment 10 from a normal state to a fault state is extracted, that is, time-series data of the nine types of features shown on the right side of Figure 3. Next, the degree of separation, which is an index showing the sensitivity when the mechanical equipment changes from a normal state to a fault state, is calculated for each of the nine types of features.

[0039] A method for calculating the degree of separation from time-series feature amount data will be described with reference to FIG. 4. Here, a method for calculating the degree of separation will be described using Fisher's linear discriminant analysis as an example. Fisher's linear discriminant analysis is a method for finding a line that can best distinguish between two sets (in this embodiment, a normal state and a fault state). At any one point on the time series, it can be determined which set the point belongs to by looking at which side the feature amount is on.

[0040] Specifically, when any point in the time series feature is used as a boundary, the data set on the left is designated A and the data set on the right is designated B, and the mean μ and variance σ of data set A and data set B are calculated. Using the calculated mean μ and variance σ of each set, the degree of separation is calculated using the following formula.

[0041]

number

[0042] For each of the nine types of feature quantities, the above process is performed while shifting the boundary between Data Set A and Data Set B along the time axis, and the degree of separation at each point on the time axis is determined. Then, for each of the nine types of feature quantities, the maximum value among the degrees of separation determined over time is determined. The larger the maximum value of separation for a feature quantity, the higher its ability to separate normal and fault conditions, i.e., the more sensitive the feature quantity is to the occurrence of a fault. Furthermore, the timing at which the maximum degree of separation is obtained for that feature quantity can be said to be the timing at which that feature quantity can best separate (distinguish) between normal and fault conditions.

[0043] Figure 9 shows an example of a work support screen that is displayed on the display unit 130 of the failure prediction device 100 when an operator selects learning data and creates a trained model through machine learning in this embodiment.

[0044] At the top left of the screen, there is a field that displays information for identifying failure cases, allowing the operator to specify, change, or confirm which failure cases' measurement data will be used as learning data.

[0045] A display field that displays the range of maximum separability values ​​is located in the upper right corner of the screen. This is to facilitate the operator's task of specifying, changing, or checking the range of maximum separability values, which is a condition for selecting features to be used in training data. The example in Figure 9 shows that the condition for selecting features to be used in training data is that the maximum separability value is 50 or greater.

[0046] In the middle section of the screen, information about the features selected that met the selection criteria is displayed, sorted from highest to lowest by maximum degree of separation. The example in Figure 9 shows that of the nine features, only three met the selection criteria: maximum current, minimum pressure, and average speed, in order of largest maximum degree of separation. For example, looking at the maximum current, it shows that the highest degree of separation, 98, was detected seven days before the occurrence of the failure. Note that in the example in Figure 9, features extracted from measurement data from different types of sensors, such as current, speed, and pressure, are selected; however, this is by chance; it is possible that features obtained by applying different processing to measurement values ​​from the same sensor may be selected. The graph column displayed on the right side of the middle section of the screen graphically displays a graph of the change over time for each feature, the timing at which the degree of separation for each feature is maximum, and the normal interval described below.

[0047] In this embodiment, data from the time-series data for the selected feature quantity during a period when the machine equipment is in a normal state (normal section) is used as learning data, and a normal section designation information field is displayed at the bottom of the screen. This field is provided for the convenience of the operator in setting the normal section and instructing confirmation and changes. In the example of Figure 9, it is indicated that the normal section is to be set to a period of 10 days or more before the occurrence of the failure.

[0048] The normal section is set as a section that is earlier (past) than any of the times when the maximum degree of separation is detected for the three selected feature quantities. If the time when the maximum degree of separation is detected for a selected feature quantity falls within the normal section, data from a fault state will be mixed into the data from the normal section for that feature quantity, which would be inconvenient for correctly machine learning the features of a normal state. In the example of Figure 9, the earliest time when the maximum degree of separation is detected among the selected feature quantities is the average speed, but this is not a problem because the normal section is set before the time when the separation of the average speed becomes maximum.

[0049] Furthermore, the earlier the normal interval is set (farther in the past) than the time of the failure, the closer it is to the initial state among normal conditions, i.e., the further away the mechanical equipment is from the failure, the more machine learning can be performed using data. However, the earlier the normal interval is set (farther in the past) than the time of the failure, the shorter the normal interval will be, and the less data can be used for machine learning. Therefore, it is desirable to set the normal interval by comprehensively taking into account the quality and quantity of the data.

[0050] The selection column displayed on the left side of the middle section of the screen is provided for the convenience of the operator to confirm that the feature has been selected and to instruct or change the selection / deselection. For example, even if the three feature values ​​mentioned above are automatically selected by specifying a selection condition of separation of 50 or more, if the operator wishes to exclude some of them at their discretion, they can do so by unchecking the selection column. Figure 10 shows an example of the display screen when the operator instructs to exclude the speed average from the features for machine learning.

[0051] In this way, according to this embodiment, a feature with high sensitivity for distinguishing between a normal state and a fault state can be selected, and data from the normal section in which the mechanical equipment is in a normal state for that feature can be extracted as data for machine learning.

[0052] The model generation button located at the right end of the lower section of the screen is an icon that allows the operator to instruct the failure prediction device to begin generating a failure prediction model through machine learning using the selected features.

[0053] Next, with reference to the schematic diagram in Figure 5, we will explain how to have an autoencoder perform machine learning using the training data extracted as described above. An autoencoder is a type of neural network that compresses (encodes) input training data into a small amount of information and then restores (decodes) it. Through machine learning, it learns "parameters for effectively compressing and restoring input data," i.e., the characteristics of the input data.

[0054] An autoencoder encodes the input value x and compresses it in the intermediate layer z. It then decodes the intermediate layer z and restores it as the output value y. In this process, it uses machine learning to minimize the restoration error J between the input and output values. That is, the autoencoder determines W and b in Equation 2 and W′ and b′ in Equation 3 so as to reduce the recovery error J in Equation 4 shown below, where s is an activation function.

[0055]

number

[0056]

number

[0057]

number

[0058] If data with similar features to the training data is input into a trained autoencoder (sometimes referred to as a trained model or failure prediction model), it will output an output value with a small restoration error through encoding and decoding using the parameters acquired during training.On the other hand, if data with different features from the training data is input into a trained model, the parameters acquired during training cannot compress and decode it properly, resulting in a large restoration error. Taking advantage of this characteristic, in this embodiment, data of the normal section of the selected feature is given to an autoencoder as an input value x to perform machine learning.

[0059] Furthermore, when performing failure prediction, the evaluation data for the same type of feature selected during learning is input to the trained model as input value x, and an output value y is output. Then, the reconstruction error of the output value y relative to the input value x is calculated, and the reconstruction error, i.e., the degree of deviation between the input and output, is treated as an index showing the degree to which the mechanical equipment deviates from its normal state.

[0060] Furthermore, in this embodiment, a judgment threshold is set in advance to be used when determining whether a mechanical equipment failure is imminent using the deviation. To set the judgment threshold, first, feature data based on actual sensor data of the mechanical equipment from a normal state until a failure occurs is input to the trained model, and the change in the deviation over time until the failure occurs is examined. As this feature data, data of the same type of feature as selected during learning (feature obtained by performing the same processing on measurement data from the same sensor) is used. Based on the change in the deviation over time, a judgment threshold is set to determine whether a failure is imminent. If the deviation is equal to or greater than the judgment threshold, it is determined that a failure of the mechanical equipment is imminent, i.e., there is a sign of a failure.

[0061] Figure 11 is a diagram specifically explaining how to determine the judgment threshold. The horizontal axis of the graph in Figure 11 is time (hour), and the vertical axis is the index value (deviation between the input and output of the trained model) indicating the degree of approaching failure, showing the change in the index value over time from the initial normal state to the occurrence of a failure (abnormal state). Assume that you want to ensure a predetermined period t between the time when the failure prediction device predicts and notifies you of an approaching failure and the time when the failure actually occurs, that is, you want the failure prediction device to predict the failure a predetermined period t before it occurs. In this case, as shown in the figure, the numerical value of the index value (deviation between the input and output of the trained model) at a point a predetermined period t prior to the occurrence of the failure is set as the judgment threshold for failure prediction (judgment threshold setting process).

[0062] We will now explain failure prediction using the above-mentioned trained model and judgment threshold. Figure 6 is a schematic diagram for explaining a failure prediction method using an autoencoder. Evaluation data indicating the operating state of the mechanical equipment at the time of evaluation is input to the trained model, and the input and output values ​​are used to calculate a deviation indicating how far the mechanical equipment is from the trained normal state. The evaluation data is data used at the time of evaluation for the same type of feature as selected during training (feature obtained by performing the same processing on measurement data from the same sensor). Specifically, as shown in FIG. 6, the evaluation data is input to the failure prediction model, and the recovery error J between the output value y and input value x of the resulting failure prediction model is calculated and treated as the deviation from the normal state. In this embodiment, this deviation is treated as an index value indicating the degree of approach to the occurrence of a failure. If the deviation (recovery error J) is equal to or greater than the determination threshold T, it is determined that the time until the occurrence of a failure is equal to or less than the predetermined period t, i.e., there is a sign of a failure. Conversely, if the deviation (recovery error J) is less than the determination threshold T, it is determined that the time until the occurrence of a failure is longer than the predetermined period t, i.e., there is no sign of a failure.

[0063] [Processing Procedure] Next, the procedure of the process executed by the failure prediction device 100 will be described with reference to the flowcharts of FIGS. FIG. 7 is a flowchart showing a processing procedure for generating a failure prediction model.

[0064] First, in step S101, the sensor data collection means 111 of the failure prediction device 100 acquires measurement data from the sensors 11 for measuring the state of the mechanical equipment 10, and stores the data in the sensor data storage means 121 (measurement data acquisition step).

[0065] Next, in step S102, the feature extraction means 112 extracts features indicating the characteristics of the operating state of the mechanical equipment 10 based on the sensor data stored in the sensor data storage means 121, and stores them in the feature storage means 122 (feature extraction process).

[0066] Next, in step S103, the control unit 110 determines whether or not a failure has occurred in the mechanical equipment 10. The occurrence of a failure may be input by a user to the control unit 110 of the failure prediction device 100 via the input unit 140, or may be input by a control unit (not shown) of the mechanical equipment 10 to the control unit 110 of the failure prediction device 100 via the network interface (NIF) 1607.

[0067] If no failure has occurred (step S103: no), the process returns to step S101, and thereafter steps S101 and S102 are repeated until a failure occurs. If a failure has occurred (step S103: yes), the process proceeds to step S104, where the separability calculation means 113 acquires the feature amounts from the feature amount storage means 122 and calculates the separability for each feature amount.

[0068] Next, in step S105, the separation degree calculation means 113 extracts the maximum separation degree value for each feature from the calculated time-series data of separation degrees, and stores the maximum separation degree value in the separation degree storage means 123. That is, the maximum separation degree value in the period from the normal state to the fault state is obtained for each of the plurality of feature amounts (maximum separation degree obtaining step).

[0069] Next, in step S106, the feature selection unit 131 obtains the maximum value of the degree of separation for each feature from the separation storage means 123, and selects the feature to be used for machine learning based on the threshold stored in the separation threshold setting means 132. Furthermore, the normal interval setting means 133 sets a normal interval of the mechanical equipment (a normal state period during which signs of failure have not yet appeared in the feature) by referring to the timing at which the maximum value of the degree of separation for the selected feature appears. Information about the selected feature and the normal interval is stored in the failure prediction model condition storage means 124 as conditions for extracting learning data for machine learning.

[0070] Next, in step S107, the data extraction means 114 extracts machine learning data from the feature storage means 122 based on the extraction conditions stored in the failure prediction model condition storage means 124, and outputs the machine learning data to the failure prediction model generation means 115. That is, data of the normal section of the selected feature is extracted and output (learning data extraction step).

[0071] Next, in step S108, the failure prediction model generation means 115 generates a trained model (failure prediction model) using the machine learning data input from the data extraction means 114 and stores it in the failure prediction model storage means 125 (trained model generation process). By performing the above series of processes, a trained model (fault prediction model) can be generated.

[0072] Next, the procedure of the process executed by the failure prediction device 100 using the generated trained model (failure prediction model) when determining whether the mechanical equipment 10 is approaching the occurrence of a failure will be described. 8 is a flowchart showing the processing procedure. The process of determining whether the mechanical equipment 10 is approaching a failure is started, for example, by a user using the input unit 140 of the failure prediction device 100 to instruct the start of the process. Alternatively, the control program of the failure prediction device 100 may be configured to automatically start the process in accordance with the operating time of the mechanical equipment 10.

[0073] When the process starts, in step S201, the sensor data collection means 111 of the failure prediction device 100 acquires measurement data from the sensors 11 for measuring the state of the mechanical equipment 10 and stores the data in the sensor data storage means 121.

[0074] Next, in step S202, the feature extraction means 112 extracts features that indicate the characteristics of the operating state of the mechanical equipment 10 based on the sensor data stored in the sensor data storage means 121, and stores them in the feature storage means 122.

[0075] Next, in step S203, the data extraction means 114 acquires information stored in the failure prediction model condition storage means 124, i.e., information on the type of feature used in machine learning, and extracts feature values ​​of the same type from the feature value storage means 122. The extracted feature values ​​are output to the failure determination means 116 as evaluation data (evaluation feature value extraction step).

[0076] Next, in step S204, the fault determination means 116 inputs the evaluation feature data input from the data extraction means 114 into the trained model (fault prediction model) stored in the fault prediction model storage means 125, and calculates the degree of deviation between the input and the output.

[0077] Next, in step S205, the failure determination means 116 compares the calculated deviation with a determination threshold value to determine whether or not the mechanical equipment 10 is approaching the occurrence of a failure, that is, whether or not there is a sign of a failure. If the deviation is equal to or greater than the determination threshold (step S205: yes), it is determined that a failure of the mechanical equipment 10 is imminent, and the process proceeds to step S206.

[0078] In step S206, the fault determination means 116 issues a notification command to the fault notification means 117. Upon receiving the notification command, the fault notification means 117 notifies the user of the determination result of the fault determination means 116. When making the notification, the user is notified via a user interface, and information related to the determination may be stored in the memory unit 120 or provided to an external device via an external interface. To notify the user, processing such as displaying on the display unit 130 of the fault prediction device 100, issuing a voice message, or printing and outputting on a medium such as paper may be performed. When notification to the user is complete, the processing ends (END).

[0079] If the deviation is less than the judgment threshold (step S205: no), it is judged that the occurrence of a failure in the mechanical equipment 10 is far away, i.e., there are no signs of a failure, and the process ends (END). Note that even if it is judged that there are no signs of a failure, the result may be notified to the user, information related to the judgment may be stored in a storage device, or provided to an external device via an external interface.

[0080] As described above, according to the embodiment, it is possible to select feature quantities that are highly correlated with the occurrence of a failure from among various feature quantities that indicate the state of mechanical equipment based on failure cases. Even if only a few failure cases are collected, it is possible to create a failure prediction model with high prediction accuracy. In other words, even with a small number of failure cases, it is possible to select feature quantities that have high separation ability for distinguishing between normal and failed states from among various feature quantities that indicate the state of mechanical equipment as data for machine learning. By selecting feature quantities that are highly correlated with the occurrence of a failure and performing machine learning, it is possible to create a trained model that is highly accurate in predicting the occurrence of a failure in advance.

[0081] [Other embodiments] The present invention is not limited to the above-described embodiment, and many modifications are possible within the scope of the technical concept of the present invention. For example, in the embodiment, a failure prediction model is created using an autoencoder by a so-called unsupervised learning method, but the present invention can also be implemented by creating a failure prediction model using a so-called supervised learning method. Supervised learning is a method of providing a large amount of data sets of certain inputs and results (labels) to a learning device, having it learn the features of those data sets, and constructing a model that estimates results from inputs, i.e., a trained model that inductively acquires the relationship between inputs and outputs.

[0082] Furthermore, in the embodiments, a method using a neural network has been described as an example of machine learning, but the machine learning method is not limited to this, and other methods such as genetic programming, functional logic programming, support vector machines, etc. Although a general-purpose computer or processor can be used as a device for performing machine learning, high-speed processing becomes possible when a graphics processing unit with a GPGPU function or a large-scale PC cluster is used.

[0083] In addition, in the embodiment, in step S106, the feature selector 131 selects the feature to be used for machine learning based on the maximum value of the degree of separation for each feature and a predetermined threshold value stored in the separability threshold setting means 132. However, the selection may be made without using a threshold value. For example, the feature values ​​may be sorted in descending order of the maximum value of the degree of separation, and the feature values ​​up to a predetermined rank (a predetermined number of feature values) may be selected for machine learning.

[0084] As described above, according to the embodiment, a failure prediction model with high failure prediction accuracy can be generated by machine learning using data on a small number of failure cases. However, this does not prohibit additional machine learning when a new failure case occurs. That is, after generating a trained model, additional learning may be performed using data on new failure cases. In this case, the selected feature quantities and judgment thresholds may be changed by referring to measurement data related to the failure case to be additionally learned.

[0085] The failure prediction device of the present invention can be applied to failure prediction of various machines and equipment, such as industrial robots, service robots, processing machines operated by computer numerical control, etc. The failure prediction device may be integrated with the machine equipment to form a failure prediction system, or the failure prediction device may be provided as part of the machine equipment.

[0086] The present invention can also be realized by supplying a program that realizes one or more functions of the embodiments to a system or device via a network or a storage medium, and having one or more processors in the computer of the system or device read and execute the program. It can also be realized by a circuit (e.g., ASIC) that realizes one or more functions. [Explanation of symbols]

[0087] 10. Mechanical equipment / 11. Sensor / 100. Failure prediction device / 110. Control unit / 111. Sensor data collection means / 112. Feature extraction means / 113. Separation degree calculation means / 114. Data extraction means / 115. Failure prediction model generation means / 116. Failure judgment means / 117. Failure notification means / 120. Storage unit / 121. Sensor data storage means / 122. Feature storage means / 123. Separation degree storage means / 124. Failure prediction model condition storage means / 125. Failure prediction model storage means / 130. Display unit / 131. Feature selection unit / 132. Separation degree threshold setting means / 133. Normal section setting means / 140. Input unit

Claims

1. An information processing method for an information processing device having a processing unit, The processing unit acquiring at least two types of measurement values ​​relating to the state of the mechanical equipment while the mechanical equipment changes from a first state to a second state; acquiring an index for a change from the first state to the second state for each of the at least two types of measurement values; selecting a predetermined measurement value from the at least two types of measurement values ​​based on the index; A period is set based on the timing when the maximum value of the index in the predetermined period becomes equal to or greater than a threshold value in the predetermined measurement value, and data in the period is used for machine learning. An information processing method comprising:

2. The processing unit a period during which the mechanical equipment was in the first state is set based on the timing at which the maximum value of the index in the predetermined period becomes equal to or greater than a threshold value in the predetermined measurement value, and data from that period is used for machine learning.

2. The information processing method according to claim 1,

3. The processing unit selecting, from the at least two types of measurement values, a measurement value at which the maximum value of the index in a predetermined period is equal to or greater than a threshold value, as the predetermined measurement value; 3. The information processing method according to claim 1, wherein:

4. The processing unit Obtaining a trained model through machine learning, and determining the state of the mechanical equipment based on the trained model and measurement values ​​relating to the state of the mechanical equipment obtained after obtaining the trained model.

4. The information processing method according to claim 1, wherein the first and second inputs are input to the first and second inputs.

5. The processing unit In acquiring the trained model, the trained model is acquired by machine learning using an autoencoder.

5. The information processing method according to claim 4.

6. The processing unit Among the timings at which the maximum value of the index in a predetermined period of time for the selected predetermined measurement value is equal to or greater than a threshold value, the predetermined measurement value in a period earlier than the earliest timing is used for machine learning.

6. The information processing method according to claim 1, wherein:

7. The index is the degree of separation, The processing unit Dividing the time series of the at least two measurement values ​​into a first data set and a second data set at at least one timing; obtaining the degree of separation based on a first mean and a first variance of the first data set and a second mean and a second variance of the second data set at each of at least one timing; 7. The information processing method according to claim 1, wherein:

8. The processing unit displaying information related to the selection of the predetermined measurement value to be used for machine learning and / or information related to the period during which the mechanical equipment was in the first state on a display unit; 8. The information processing method according to claim 1, wherein:

9. The processing unit displaying on a display unit information relating to the maximum value of the index in a predetermined period, and / or information relating to the timing at which the maximum value of the index in a predetermined period becomes equal to or greater than a threshold value, and / or information relating to the designation of the period during which the mechanical equipment was in the first state; 9. The information processing method according to claim 1, wherein:

10. The processing unit notifying the result of the determination of the state of the mechanical equipment; 10. The information processing method according to claim 1, wherein:

11. An information processing method for an information processing device having a processing unit, The processing unit acquiring at least two types of measurement values ​​relating to the state of the mechanical equipment while the mechanical equipment changes from a first state to a second state; acquiring an index for a change from the first state to the second state for each of the at least two types of measurement values; using a predetermined measurement value selected from the at least two types of measurement values ​​based on the index for machine learning; a screen for allowing a user to set a threshold value for the index when selecting the predetermined measurement value is displayed on a display unit; An information processing method comprising:

12. An information processing method for an information processing device having a processing unit, The processing unit acquiring at least two types of measurement values ​​relating to the state of the mechanical equipment while the mechanical equipment changes from a first state to a second state; acquiring an index for a change from the first state to the second state for each of the at least two types of measurement values; using a predetermined measurement value selected from the at least two types of measurement values ​​based on the index for machine learning; the predetermined measurement values ​​are displayed on the display unit in descending order of the maximum value of the index in a predetermined period of time; An information processing method comprising:

13. An information processing method for an information processing device having a processing unit, The processing unit acquiring at least two types of measurement values ​​relating to the state of the mechanical equipment while the mechanical equipment changes from a first state to a second state; acquiring an index for a change from the first state to the second state for each of the at least two types of measurement values; displaying a screen on a display unit that allows a user to select whether or not a predetermined measurement value selected from the at least two types of measurement values ​​based on the index is to be used for machine learning; using the predetermined measurement value selected by the user for machine learning; An information processing method comprising:

14. The first state is a state in which the mechanical equipment is normal, and the second state is a state in which the mechanical equipment is abnormal.

14. The information processing method according to claim 1,

15. The processing unit extracting data relating to at least two feature quantities from the at least two types of measurement values; acquiring the index for the data relating to the at least two feature amounts, and selecting predetermined data from the data relating to the at least two feature amounts based on the index, thereby selecting the predetermined data as the predetermined measurement value; a period is set based on the timing when the maximum value of the index in the predetermined period becomes equal to or greater than a threshold value in the predetermined data, and the data in the period is used for machine learning; 5. The information processing method according to claim 4.

16. The processing unit Inputting data of the same type as the selected predetermined data, from the time when the machine equipment moves from the first state to the second state, into the trained model; Calculating the degree of discrepancy between input data input to the trained model and output data output from the trained model; setting a judgment threshold based on a change in the deviation degree over time during a period when the mechanical equipment changes from the first state to the second state; acquiring, as an evaluation feature, data of the same type as the selected predetermined data using a measurement value related to the state of the mechanical equipment at the time of evaluation; In determining the state of the mechanical equipment, an index value indicating the degree to which the mechanical equipment has deviated from the first state is obtained using the evaluation feature and the trained model, and the index value and the determination threshold value are used to determine the state of the mechanical equipment at the time of the evaluation.

16. The information processing method according to claim 15.

17. The processing unit selecting a predetermined number of measurement values ​​from the at least two types of measurement values ​​in descending order of the maximum value of the index in a predetermined period as the predetermined measurement values; 17. The information processing method according to claim 1.

18. An information processing device having a processing unit, The processing unit acquiring at least two types of measurement values ​​relating to the state of the mechanical equipment while the mechanical equipment changes from a first state to a second state; acquiring an index for a change from the first state to the second state for each of the at least two types of measurement values; selecting a predetermined measurement value from the at least two types of measurement values ​​based on the index; A period is set based on the timing when the maximum value of the index in the predetermined period becomes equal to or greater than a threshold value in the predetermined measurement value, and data in the period is used for machine learning.

1. An information processing device comprising:

19. An information processing device having a processing unit, The processing unit acquiring at least two types of measurement values ​​relating to the state of the mechanical equipment while the mechanical equipment changes from a first state to a second state; acquiring an index for a change from the first state to the second state for each of the at least two types of measurement values; using a predetermined measurement value selected from the at least two types of measurement values ​​based on the index for machine learning; a screen for allowing a user to set a threshold value for the index when selecting the predetermined measurement value is displayed on a display unit; 1. An information processing device comprising:

20. An information processing device having a processing unit, The processing unit acquiring at least two types of measurement values ​​relating to the state of the mechanical equipment while the mechanical equipment changes from a first state to a second state; acquiring an index for a change from the first state to the second state for each of the at least two types of measurement values; using a predetermined measurement value selected from the at least two types of measurement values ​​based on the index for machine learning; the predetermined measurement values ​​are displayed on the display unit in descending order of the maximum value of the index in a predetermined period of time; 1. An information processing device comprising:

21. An information processing method for an information processing device having a processing unit, The processing unit acquiring at least two types of measurement values ​​relating to the state of the mechanical equipment while the mechanical equipment changes from a first state to a second state; acquiring an index for a change from the first state to the second state for each of the at least two types of measurement values; displaying a screen on a display unit that allows a user to select whether or not a predetermined measurement value selected from the at least two types of measurement values ​​based on the index is to be used for machine learning; using the predetermined measurement value selected by the user for machine learning; An information processing method comprising:

22. A production system comprising: the information processing device according to any one of claims 18 to 20; and the machinery and equipment.

23. A method for manufacturing an article, comprising manufacturing the article using the production system according to claim 22.

24. A program that enables a computer to execute the information processing method according to any one of claims 1 to 17 or claim 21.

25. A computer-readable recording medium storing the program according to claim 24.

Images